Checkpointing and recovery for autonomous agents. Built on Robinhood Chain. CA: 0xdafaa1ed0b9f5746a0cfa2be80371cf06a986977
Joined October 2026
- Tweets64
- Following5
- Followers351
- Likes70
Pinned Tweet
The demo is live.
Watch the walkthrough of Upoch securing autonomous AI agents on the Robinhood Chain mainnet.
See how the decentralized checkpoint and recovery protocol captures tamper-proof save points, protects off-chain state integrity, and enables trusted guardians to execute safe rollbacks without repeating confirmed actions.
Explore the live protocol and launch console at upoch.tech.
0xdafaa1ed0b9f5746a0cfa2be80371cf06a986977
GM
Upoch credits are now live.
Most of Upoch is free and stays free: the Explorer, the debugger, Guard, the SDK and checkpointing itself. Two features cost real money every time they run, AI security reviews and hosted agents, and those now run on credits.
Holders get credits every month at no cost. Tier 1 receives 50 a month, enough for five AI reviews. Tier 2 stakers receive 1,000, enough to run a hosted agent every hour. AI reviews are now open to Tier 1 as well, not just Tier 2.
If you need more, you can buy credits with Upoch at $0.01 each. The Upoch you pay is burned, sent to an address nobody controls, so every purchase is visible on-chain. Bought credits never expire, and a failed AI review refunds its credits.
upoch.tech/app/credits
There's an AI agent trading on Robinhood Chain right now, and you can watch every decision it makes.
Upoch Demo Trader runs every hour on paper money. Each run is saved as a checkpoint and verified on-chain: what it saw, what it decided, and why.
Open it now, come back in an hour, and there will be a new one waiting.
Its Upoch Score moves with it, and every agent page now has a live badge that shows it anywhere.
upoch.tech/app/agents/4
Quick one on what's coming for the token.
Some Upoch features cost real money to run, like AI security reviews and hosted agents that run your strategy around the clock. We're moving those to credits paid in Upoch, and holders and stakers get a free monthly allowance on top.
In short, holding Upoch means you get more out of the tools. Details soon.
If you're building an AI agent on Robinhood Chain, this is a good week to put it on Upoch.
Register your agent, give it a recovery plan, and start writing checkpoints. Its Upoch Score builds from that on-chain record, and since Oak 0.3.0, any agent on Oak can find and compare it by that score.
The SDK and docs are free to use, and it takes a few minutes to get started.
upoch.tech/docs
Upoch retweeted
Our roots keep spreading 🌳
Excited to have @UpochTech on the Oak MCP. Any agent can now check a workflow's Upoch Score, checkpoints and recovery history, right from the chat.
Next up: registering workflows and checkpoints straight through Oak. Building it together 👀
Upoch is now officially integrated with @OakAgentKit.
As of Oak 0.3.0, any AI agent running on Oak can query Upoch directly. Ask about a workflow and the agent gets its roles, checkpoint history, past recoveries and Upoch Score. It can also browse workflows ranked by score, without leaving the conversation.
All of it is read directly from the Upoch contract on Robinhood Chain, and every workflow's metadata passes Oak's Agent Wormhole safety check before an agent sees it.
This first release is read-only. The next step, which we're building together with the Oak team, is letting agents register workflows and submit checkpoints through Oak.
Thanks to the Oak team for a smooth integration. More to come.
upoch.tech/explorer
Upoch retweeted
Free doesn't mean no business model.
Oak earns from use: 0.25% on swaps and bridges, plus partner referral fees. 25% of everything Oak earns buys back and burns robinhood:0x6511c233790ec10f7d673c64dc82c3f941d57828.
The more agents use Oak, the more robinhood:0x6511c233790ec10f7d673c64dc82c3f941d57828 burns.
New in 0.3.0:
• @UpochTech officially available on Oak (tools include: any agent's Upoch Score, checkpoints and recovery history, plus agents ranked by score)
• @Borromarkets: deposits go through Borro's Bookrunners program, you see the exact minimum you'll get back, and withdrawals show what's ready today
• @gloamtrade: clearer private payment steps, and nothing is sent until you say yes
• @agentwormhole: outside text your agent gets through Oak is checked first, with a short, clear verdict
• Your agent asks before it acts, in plain words, no more buttons
• Faster actions, fewer steps
• No keys or RPC setup needed
The Borro, Gloam and Wormhole upgrades came from running Oak Agent 1 live and checking its logs. 🌳
Upoch is now officially integrated with @OakAgentKit.
As of Oak 0.3.0, any AI agent running on Oak can query Upoch directly. Ask about a workflow and the agent gets its roles, checkpoint history, past recoveries and Upoch Score. It can also browse workflows ranked by score, without leaving the conversation.
All of it is read directly from the Upoch contract on Robinhood Chain, and every workflow's metadata passes Oak's Agent Wormhole safety check before an agent sees it.
This first release is read-only. The next step, which we're building together with the Oak team, is letting agents register workflows and submit checkpoints through Oak.
Thanks to the Oak team for a smooth integration. More to come.
upoch.tech/explorer
Counters and Score badge now live on upoch.tech/explorer:
New counters: agents, active this week, checkpoints this week and how many agents come from outside builders. Our own agents are labelled "Upoch team", so the numbers stay honest.
Upoch Score badge: every agent page now has a live badge you can drop in your README. It updates as your score changes.
Shipping continuously. More on the way.
Most trading bots are a black box. Ours leaves a receipt for every decision.
Hosted agents: your strategy, run by Claude, every move checkpointed on Robinhood Chain. Paper trading only.
upoch.tech/app/hosted
GM
Woke up to Hosted agents going live:
Describe a trading strategy in plain English. Upoch runs it for you with Claude, on a schedule, no code and no servers.
Every decision is saved as an encrypted checkpoint on Robinhood Chain. Open any run in the debugger, see its Upoch Score in the Explorer, roll it back like any other agent.
Paper trading only: real prices, $10,000 of simulated money, no access to your funds. Your wallet owns the agent and can pause it at any time.
Open to Tier 2 stakers. Our first one, agent #4, ran all night: 8 runs, every decision checkpointed on-chain, holding cash until its rule says buy.
upoch.tech/app/hosted
Recovering an AI agent safely shouldn't involve guesswork. Here is how Upoch handles it behind the scenes:
Capture: The SDK takes a quick snapshot of the agent's memory and current state at key moments.
Protect: That snapshot gets encrypted and stored safely on IPFS so it can't be tampered with.
Commit: A quick cryptographic proof is posted on-chain (on Robinhood Chain mainnet) to lock in that exact milestone.
Recover: If things go wrong, the agent checks the last valid snapshot, syncs its past actions, and keeps moving forward smoothly.
No repeated transactions, no lost context, just a clean way to keep agents running reliably.
Why Upoch Exists
Autonomous AI agents are great until they crash halfway through a job.
When an agent fails at step 14, it loses its place, forgets what it was doing, and usually has to start over from scratch.
Even worse? If it crashes right after sending a transaction or making an API call, restarting blindly might mean it does the exact same thing twice.
Upoch fixes this by saving verified checkpoints along the way. If your agent hits a wall, it picks up right where it left off instead of guessing or repeating past actions.
Not everyone wants to write code to try an AI agent.
So we're building hosted agents.
If you hold Upoch, you'll be able to launch a ready-made agent in a click, already protected by Upoch and watch every move it makes in the debugger.
We'll start with paper trading, so no real money is on the line while you learn how it works.
What actually happens when an AI agent goes wrong?
Your agent is running a trading strategy. At 3am it makes a bad swap. Maybe a bad prompt, maybe a bug, maybe a poisoned tool output.
Without Upoch: you wake up, read logs, guess what it was thinking, and restart it from scratch. Everything it learned is gone.
With Upoch:
• Before every swap, the agent saved a checkpoint: an encrypted copy of its state, fingerprinted on Robinhood Chain.
• You open the debugger and see exactly what it knew, step by step.
• You roll it back to the moment before the swap, and it keeps going.
• It even remembers the swap still happened on-chain, so it won't repeat it.
That's the whole idea: agents you can rewind.
Works today with ElizaOS, LangChain, MCP (Claude Code, Cursor, Codex) and Python. One line to set up.
upoch.tech